Hello everybody

What's the difference between structured mesh and unstructred mesh? Thanks.

Andrew

Andrew:

Typically a structured mesh is comprised of hex (brick)elements (quads in 2D) that follow a unifrom pattern. An unstructured mesh does not follow a unifrom pattern, usually comprised of tet elements (tris in 2D).

derrek

Hi Andrew,

A single block structured mesh may comprise of square elements (2D) or hexahedral elements (3D) which are orthogonal in i, j space (2D) or i, j, k space (3D). Although is is also possible to have wedges (3D), triangles (2D) and pyramids (3D) in a structured mesh.

Looking at a 2D example, for simplicity, every node in a 2D structured mesh has a corresponding integer i and j index value which is unique. The physical locations of the nodes are stored in a table or are functionally related to the mesh space (ie (x,y)= f(i,j)). It is also implied then that the neigbours of node (i,j) are (i-1,j), (i+1,j), (i,j-1), (i,j+1), (i-1,j-1), and (i+1,j+1). If you are writing a CFD code, a structured mesh makes it very easy to loop through neighbours and can be efficient with memory.

A structured mesh has many coding advantages, but it may be difficult to conform a single block to a complicated shape. Code developers have got around this by allowing multiple blocks (multiblock unstructured), but this can make the internal memory strucutres more inefficient. Another way to make the mesh generation simpler, and improve code performance in other ways is to throw away the block structure and replace indices with node numbers and a connectivity table. This is known as an unstructured mesh, because it lacks the i,j,k structure.

A common misperception is that a hex mesh is strucutred and a tet mesh is unstructured. It is accurate to say that a tet mesh is unstructured, however a hex mesh can also be unstructred. The difference between a structured hex mesh and an ustructured hex mesh is simply in how the data is stored.

Regards, Robin

Robin:

Thanks for the thorough explanation. I should have used the word "typically" a little more forcefully. I didn't mean to imply a definition of structured & unstructured.

derrek

[Jaiganesh S]

1.What is the main different between in structured hexa and unstructured hexa?

[ghorrocks]

Structured hex meshes are in a grid, so an element can be identified by and ijk coordinate. Unstructured hex meshes are, well, unstructured.

[heidar]

hey guys
In fact a mesh is said to be structured if one can found a primitive motif which is regular in space. This implies that for instance node connectivities has a fixed pattern. But the geometry may be not constant, for instance your mesh can be deformed from place to place (for instance to match a specific boundary). The only thing you know for sure when the mesh is regular is that the node (or elements) connectivty is always the same. This may be useful for algorithm design since you always have the same pattern.
However unstructured mesh are generally more flexible, since the connectivity pattern is not fixed, at the cost of having to store the connectivity. There is no generic answer to your last question, since it depends strongly upon the application you have in mind.

[masoudmohammadian]

Could you please explain it more?

regards

[ghorrocks]

This topic is covered in any CFD text book. I do not fancy retyping text books into forums.

[Abhya]

Originally Posted by ghorrocks

This topic is covered in any CFD text book. I do not fancy retyping text books into forums.

I hope you realize we dont want a retyped Text explanation...
Something like this (quoted below - post by Robin) is much better than that..I'd call that condensed understanding ... and that's what guys mean while asking such questions .. they wish to save some time and effort by learning from a well learned fellow..!

Originally Posted by Robin
;65829

Hi Andrew,

A single block structured mesh may comprise of square elements (2D) or hexahedral elements (3D) which are orthogonal in i, j space (2D) or i, j, k space (3D). Although is is also possible to have wedges (3D), triangles (2D) and pyramids (3D) in a structured mesh.

Looking at a 2D example, for simplicity, every node in a 2D structured mesh has a corresponding integer i and j index value which is unique. The physical locations of the nodes are stored in a table or are functionally related to the mesh space (ie (x,y)= f(i,j)). It is also implied then that the neigbours of node (i,j) are (i-1,j), (i+1,j), (i,j-1), (i,j+1), (i-1,j-1), and (i+1,j+1). If you are writing a CFD code, a structured mesh makes it very easy to loop through neighbours and can be efficient with memory.

A structured mesh has many coding advantages, but it may be difficult to conform a single block to a complicated shape. Code developers have got around this by allowing multiple blocks (multiblock unstructured), but this can make the internal memory strucutres more inefficient. Another way to make the mesh generation simpler, and improve code performance in other ways is to throw away the block structure and replace indices with node numbers and a connectivity table. This is known as an unstructured mesh, because it lacks the i,j,k structure.

A common misperception is that a hex mesh is strucutred and a tet mesh is unstructured. It is accurate to say that a tet mesh is unstructured, however a hex mesh can also be unstructred. The difference between a structured hex mesh and an ustructured hex mesh is simply in how the data is stored.

Regards, Robin

Super clear explanation btw Robin

[ghorrocks]

Yes, Robin's answer was good and he should be commended for that. But vague questions like "Could you please explain it more?" really don't encourage people to write thoughtful answers as there was little thought put into the question. You will find that to get good answers you need to ask good questions - which means clearly stating what bit you don't understand and explaining what background reading/knowledge you have in the topic.

[majid fahim]

download (1).jpg

download.jpg

[Sumanth_094]

Hello guys,

sorry for reviving this almost dead Thread. I am currently performing simulations for film cooling of gas turbine blades and I have two sets of simulations performed with structured and the unstructured meshes. Since unstructured meshing is faster (atleast for me), its preferred over structured meshing but ofcourse it comes with a price and thats accuracy. I have a simulation result perfomed with a Structured mesh with a mesh density of 4.4 million nodes and to acheive 'similar' results with an unstructured mesh, I would be needing a mesh with 36 million nodes. Thats quite high and it costs a lot more computational power and time. I use ICEM CFD for meshing. Mesh study was done for both the types of mesh and 4.4mil structured mesh was already giving good results and further refinement wasn't needed. The mesh study for unstructured meshing was done keeping in mind the results of structured mesh and as I said, we needed a high density mesh.

Is there anyway in which one could bring down the computational time of this unstructured mesh without the accuracy of the result taking a hit? By that I mean acheiving a low density tetra mesh which could still deliver matching results with my structured mesh of 4.4million nodes.

Thanks a lot for your time!

Best regards,
Sumanth

[Gert-Jan]

Don't think that you can bring down the computational time. I assume that if it would be possible, than it would be there already .

But maybe....... you can create a mix of both worlds in ICEM by converting your tet mesh to a hex mesh, where possible. Did you ever try that?
It is an option in edit mesh. There select to convert 12 tets into 1 hex, with a quality criteria of 0.2. By lowering this value, you can let ICEM convert more tets to hex.
Since the best quality (>0.2) tet elements are in the bulk of your domain, these elements will be converted predominantly. You will end up with a transition region with the worst tets. These are located between your prisms at the wall, and hex elements in the bulk.

[Sumanth_094]

Originally Posted by Gert-Jan

Don't think that you can bring down the computational time. I assume that if it would be possible, than it would be there already .

But maybe....... you can create a mix of both worlds in ICEM by converting your tet mesh to a hex mesh, where possible. Did you ever try that?
It is an option in edit mesh. There select to convert 12 tets into 1 hex, with a quality criteria of 0.2. By lowering this value, you can let ICEM convert more tets to hex.
Since the best quality (>0.2) tet elements are in the bulk of your domain, these elements will be converted predominantly. You will end up with a transition region with the worst tets. These are located between your prisms at the wall, and hex elements in the bulk.

Thanks a lot for the super fast reply
Thats one of the options which I was considering but haven't implemented yet. I wonder how the mesh turns out to be. If its as clean as the mesh which one would achieve with blocking, that would already save me a lot of time.
I am performing that as I type this. Lets see
Thanks a lot once again!

[ghorrocks]

If there was a simple way of making a tet mesh more accurate without refining the mesh, don't you think that would be made the default process?

The simplest way of making a large tet mesh simulation run faster is to get more parallel nodes or a more powerful computer.

[Sumanth_094]

Originally Posted by ghorrocks

If there was a simple way of making a tet mesh more accurate without refining the mesh, don't you think that would be made the default process?

The simplest way of making a large tet mesh simulation run faster is to get more parallel nodes or a more powerful computer.

Well, I aint just referring to the process of refinement. I was working with ANSA for the last 7-8 months and have resumed again with ICEM. ANSA offered a wide range of meshing algorithms to obtain an unstructured mesh (ICEM also has a couple of them, I know) but I aint an experienced ANSA user and I never thought about doing a structured vs unstructured comparison with that either.
As of now, I use Robust octree to get a surface mesh in ICEM and then use delaunay for the volume meshing and then proceed with prisms.

I wanted to know if the meshing algorithm/procedure has got any role to play here too. I never did that study myself but before proceeding down that road, I wanted to know if anyone here actually tried doing that.

Also, running a 36M mesh isn't a problem since the simualtions are run on a cluster and I have the result in less than a days time but trying to optimize is always good, or?

Btw, thanks for your time Glenn. Your replies have helped me learn a lot and I find your inputs always valuable!

[ghorrocks]

Also note the effect of the mesh depends on what you are modelling. If you are modelling a single phase low Re flow then you can use meshes of terrible quality and they will still work fine. But if your simulation has shock waves, multiphase (especially free surface with surface tension) or other complex physics your mesh quality requirements increase drastically. So there is not a universal answer to your question.

Also keep in mind that for commercial use - time is money. So if a simple tet mesh with little geometry modification gets you results which are good enough (even if it takes a big cluster to run it); that may be better than paying for a CFD expert to spend hours and hours cleaning up the geometry so you can use a higher quality mesh strategy. So keep meshing effort to what the simulation requires.

[Sumanth_094]

I am dealing with the former. The physics behind my model isn't that complicated. Its a single phase flow. But its crazy that even after so much refinement of my tetra mesh, some details still werent captured like in Hex. But now, I understand the difference also a bit better

Also, you are absolutely right on that part. Time is indeed more precious. I am doing this for my thesis and I had more freedom to explore some of the options and this question just popped up few days ago in my mind and I thought I would dig into it more during my free time

[ghorrocks]

Have a look in the documentation on the numerical approach. The solver forms control volumes around the nodes, which means for tet meshes the control volumes are polygonal. If you use a structured mesh the control volumes are also hexes, and that makes the control volumes have less faces (so less fluxes to keep track of, less memory) and the control volume faces can potentially be closer to perpendicular which helps reduce numerical diffusion. Hope that gives you some clues for some things to research on this topic.